Current research projects

• Learning Trustworthy Planning Algorithms (LTPA)
• Collaborative Constraint-Based Planning (CCBP)
• Neuro-Symbolic AI for Improving Energy Efficiency in 6G (AI6G)
• Robust Planning with Large Language Models (LLMPlan)
• AI for Attack Identification, Response and Recovery (AIR²)
• Parallel AI Planning (PAIP)
• Model-Based Attention for Scalable AI Planning (MBASAP)
• Online Learning For AI Planning (OLP)
• Learning Reliable Algorithms for AI Planning

Previous research projects

• Symbolic Search for Diverse Plans and Maximum Utility (SSDPMU)
• Learning Dynamic Algorithms for Automated Planning (LDAAP)
• Representation Learning for Acting and Planning (RLeap)

General research topics

Within our lab, we develop machines that can reason and act in complex environments. Our primary research area is automated planning, which we complement with techniques from machine learning, combinatorial optimization and operations research. Our main topics of interest include:

• Theory of Planning: We contribute to the theoretical foundations of Automated Planning, studying the complexity and expressiveness of planning problems and algorithms.
• Efficient Planning Algorithms: We design and implement scalable planning algorithms based on heuristic state-space search, symbolic search and decoupled search and SAT-based approaches.
• Learning Planning Models: We develop algorithms that extract the dynamics of an observed environment to learn compact descriptions of planning tasks.
• Generalized Planning: We create methods for learning how to solve a whole class of tasks efficiently, including general policies represented as sketches, graph neural networks and transformers.
• Planning and Reinforcement Learning: We combine the interpretability of planning with the flexibility of reinforcement learning.
• Planning and Large Language Models: We bridge symbolic planning and LLMs by translating natural-language tasks into formal planning problems and using language models to generate heuristics.
• Explainable Planning: We develop methods for explaining plans and policies by exploring plan spaces interactively and providing counterfactual reasoning over agent behavior.
• Planning Beyond the Classical Setting: We extend planning to numeric and probabilistic settings, and apply planning techniques to problems such as algorithm discovery and multiple sequence alignment.

In summary, we strive to create AI systems that efficiently solve intricate sequential decision-making problems, based on solid theoretical foundations and practical algorithms.

Acknowledgments

Our research is supported by the Wallenberg AI, Autonomous Systems and Software Program (WASP), generously provided by the Knut and Alice Wallenberg Foundation (KAW). We also acknowledge support from the Swedish Foundation for Strategic Research (SSF) and the Swedish Research Council (VR).